Every company has one: the person who simply knows. The service manager who can diagnose a system over the phone from a homeowner's description of a noise. The CSR who can tell which callers will book before they finish their first sentence. The estimator whose numbers are somehow always right. We call them subject-matter experts, and for the entire history of business, their career path was linear: know more, become more valuable, become harder to replace.
AI inverts that path. I call it the SME Inversion, and it is the single most important career conversation you can have with your best people this year.
The inversion, in three acts
Act one: the expert. You know the thing. Your value is doing the thing and being the person everyone comes to about the thing. This is where almost everyone is today — and it is quietly becoming the most exposed position in the building, because the thing you know is exactly what machine intelligence learns to do first.
Act two: the engineer of your own replacement. The expert's knowledge gets encoded — into prompts, workflows, decision rules, training data, automations. Somebody is going to do that encoding. The inversion says: it should be the expert themselves. The person who knows the work is the only person who can teach a system to do it well, catch where it's wrong, and know which edge cases will burn a customer. An expert encoded badly by an outsider becomes a bad system. An expert who encodes themselves becomes the author of the system.
Act three: the steward. Once the system runs, the author doesn't disappear — they graduate. They monitor it, correct it, improve it, and handle the exceptions it escalates. One person's judgment, which used to process one call at a time, now supervises every call the system touches. Same knowledge. Radically larger footprint.
"Train your replacement" used to be a threat
I understand the flinch. For generations, training your replacement meant you were on the way out. The inversion only works if ownership makes it true that encoding yourself is a promotion — in title, in pay, in status. Nobody automates themselves out of a job they are paid to keep; they will slow-walk it, hedge it, and quietly sabotage it, and they will be rational to do so. The org design has to come first:
- The person who encodes a workflow owns the system that runs it.
- Compensation follows the system's output, not the hours the person used to bill.
- The steward role is explicitly senior to the doer role it replaced — because it is: one steward's judgment now covers the volume of many doers.
Why the trades will do this better than the office towers
The trades have run knowledge transfer for centuries: it's called apprenticeship. Master to journeyman to apprentice — the expert teaching their replacement is not a threat in this industry; it is the tradition. The only new part is that one of the apprentices is now made of software. The industries that will struggle with the inversion are the ones where expertise was hoarded as job security. The trades already have the cultural muscle; they just need the career path drawn.
The question to ask your best person this week
"If we built a system that could do sixty percent of what you do, would you rather be the person it replaced or the person who built and runs it?" Every ambitious person picks the second answer — and the moment they do, your retention problem and your automation roadmap become the same project. That is the inversion working for you instead of against you. It's also exactly how my own company runs now: the 38 people at CI Web Group are the stewards of the engine that replaced the work of 320.
I wrote Hands Up for the generation that will inherit what we build — and the SME Inversion is a large part of what they're inheriting: companies where the most respected people are not the ones who guard what they know, but the ones who taught it to a system and leveled up. The essay series that follows this one covers the mechanics — job descriptions for the token era, incentive design, and what stewardship looks like day to day.